In order to improve the accuracy of respiratory signal to distinguish driving fatigue, the relationship was explored between respiratory signal and driver fatigue state through simulated driving test. The concept of respiratory fatigue node was proposed, and the driver's fatigue state was distinguished based on the respiratory fatigue node. Firstly, the respiratory signals of drivers were collected by simulated driving test, and the Karolinska sleepiness scale (KSS) was used to quantify the subjective self-evaluation of their fatigue degree. Secondly, the percentage of eyelid closure over the pupil over time (PERCLOS) was used as a reference, combined with subjective self-evaluation feedback to calibrate the respiratory fatigue nodes of drivers. Finally, based on the respiratory fatigue nodes, the random tree (RT) algorithm was used to obtain the discriminant model of mild/severe respiratory fatigue change nodes. The results show that the model can identify the driver's respiratory fatigue state more timely and accurately and the accuracy of the screening conditions based on the random tree algorithm for the identification of mild respiratory fatigue change nodes is higher than that of severe respiratory fatigue change nodes. The average recognition errors of mild/severe respiratory fatigue change nodes are 3.50 min and 3.66 min, respectively, the prediction accuracy is 92.09% and 92.03% respectively.